GraphQL Field Discovery & Query Generation with Generative AI
summary
Intuit ingests a supergraph of millions of lines by splitting every GraphQL field into field chunks, embedding them, storing them in AWS OpenSearch and retrieving top-k results with LangChain before running RAG with GPT‑4. A dynamically generated minimal schema (~8k token) is sent to the LLM to produce executable GraphQL or a ranked list of attributes, cutting developer discovery from 2–3 weeks to 20–30 minutes. Tooling includes collision detection, schema enrichment, human + synthetic golden sets, and precision/recall evaluation; end‑to‑end latency is typically a few to ~12 seconds.
AI-written summary · about
chapters (7)
- 0:00Introduction and team intro
- 1:28Customer problem and challenges
- 4:21Gen AI fundamentals and RAG
- 10:03Field discovery and query generation
- 13:58Evaluation and metrics
- 17:33Schema enrichment and improvements
- 26:26Q&A session
related talks
15mTraining and Using Generative AI Models with GraphQL
24mFrom Query to Conversation: GraphQL as an AI Interface Layer
26mJust Ask: Building a Natural Language GraphQL Agent from the Graph You Already…
31mGraphQL Embeddings: AI-Powered Dynamic Operations From Schema To IDE
26mEvolving GraphQL: Designing APIs That Deliver Insights with Intelligent Resolv…
31mWhy AI needs GraphQL
27mHow GraphQL Gives Coding Agents Eyes on What Actually Matters
29mClosing the Loop: How GraphQL Gives Coding Agents Eyes on What Actually Matters